Environmental Preferences and Technological Choices : Is ...Dirty dirty + grey dirty+ grey + other...

36
Environmental Preferences and Technological Choices : Is Market Competition Clean or Dirty? * Philippe Aghion - Roland B´ enabou - Ralf Martin § - Alexandra Roulet January 5, 2021 Abstract We investigate the effects of consumers’ environmental concerns and market competition on firms’ decisions to innovate in “clean” technologies. Agents care about their consumption and environmental footprint; firms pursue greener prod- ucts to soften price competition. Acting as complements, these forces determine R&D, pollution, and welfare. We test the theory using panel data on patents by 8,562 automobile-sector firms in 41 countries, environmental willingness-to- pay, and competition. As predicted, exposure to prosocial attitudes fosters clean innovation, all the more so where competition is strong. Plausible increases in both together can spur it as much as a large fuel-price increase. * We are thankful for comments and suggestions from Ufuk Akcigit, Gene Grossman, Steve Redding, Jean Tirole, John Van Reenen, and from participants in the Institutions, Organizations and Growth (IOG) group at the Canadian Institute for advanced Research, at the EBRD Confence on “Envi- ronmental Economics and the Green Transition” and at the TSE workshop on “Markets, Morality and Social Responsibility”. L´ eo Aparisi de Lannoy provided superb research assistance. Aghion and enabou gratefully acknowledge financial support from the Canadian Institute for Advanced Study, and B´ enabou from the Innovation Lab at College de France as well. Coll` ege de France, CEP & CEPR Princeton University, NBER, CEPR, Briq, ThreD, IZA and ERINN. § Imperial College London, CEP & CEPR INSEAD and CEPR 0

Transcript of Environmental Preferences and Technological Choices : Is ...Dirty dirty + grey dirty+ grey + other...

  • Environmental Preferences and Technological Choices :

    Is Market Competition Clean or Dirty?∗

    Philippe Aghion† - Roland Bénabou‡ - Ralf Martin§ - Alexandra Roulet¶

    January 5, 2021

    Abstract

    We investigate the effects of consumers’ environmental concerns and market

    competition on firms’ decisions to innovate in “clean” technologies. Agents care

    about their consumption and environmental footprint; firms pursue greener prod-

    ucts to soften price competition. Acting as complements, these forces determine

    R&D, pollution, and welfare. We test the theory using panel data on patents

    by 8,562 automobile-sector firms in 41 countries, environmental willingness-to-

    pay, and competition. As predicted, exposure to prosocial attitudes fosters clean

    innovation, all the more so where competition is strong. Plausible increases in

    both together can spur it as much as a large fuel-price increase.

    ∗We are thankful for comments and suggestions from Ufuk Akcigit, Gene Grossman, Steve Redding,

    Jean Tirole, John Van Reenen, and from participants in the Institutions, Organizations and Growth

    (IOG) group at the Canadian Institute for advanced Research, at the EBRD Confence on “Envi-

    ronmental Economics and the Green Transition” and at the TSE workshop on “Markets, Morality

    and Social Responsibility”. Léo Aparisi de Lannoy provided superb research assistance. Aghion and

    Bénabou gratefully acknowledge financial support from the Canadian Institute for Advanced Study,

    and Bénabou from the Innovation Lab at College de France as well.

    †Collège de France, CEP & CEPR‡Princeton University, NBER, CEPR, Briq, ThreD, IZA and ERINN.§Imperial College London, CEP & CEPR¶INSEAD and CEPR

    0

  • 1 Introduction

    Should private firms get involved in mitigating climate change? A traditional view

    against such corporate activism is that firms should concentrate on maximizing prof-

    its, and let governments deal with externalities. In practice, however, we often see

    governments being ineffective at addressing environmental problems.1 It then falls

    upon intrinsically motivated consumers, investors and firms to “do their part” through

    other channels.

    This paper shows how citizens’ social-responsibility concerns and the degree of compe-

    tition between firms jointly shape the direction of innovation, acting as complements.

    We first develop a simple model of innovation where agents care about both the level

    and the environmental footprint of their consumption. We analyze how these “ethical”

    preferences, together with market structure, affect the equilibrium amount of clean

    R&D, and through it aggregate pollution and welfare.

    While the direct, short-run impact of competition on the environment is always negative

    –lower prices induce more consumption and therefore more pollution2– it can also

    encourage clean innovation as a means of product differentiation. Intuitively, firms

    will seek to develop greener products when facing more environmentally motivated

    customers, and the more so, the harder they must compete for them.

    Due to its offsetting quantity and quality effects, the impact of competition on emis-

    sions has a concave profile. Furthermore, because social responsibility and competition

    leverage each other, when the former is strong enough the profile can be hump-shaped,

    or even decreasing, reversing the direct effect. Similarly, more prosocial consumers not

    only push this profile down, but also make increases in competition (desirable for the

    usual reasons) less environmentally costly, or even beneficial.

    In the second part of the paper, we bring together patent data, survey data on environ-

    mental values, and competition measures to test the model’s key comparative statics.

    We relate the extent to which firms innovate in a clean direction to firm-specific mea-

    sures of exposure to pro-environmental attitudes and competition. Our data covers

    8,562 firms and 41 countries during 1998-2002 and 2008-2012, with around 100,000

    1Bénabou and Tirole (2010) discuss the sources of these limitations or failures, and how they createa scope for individual and corporate social responsibility.

    2The examples of China or India today, or of the increasing market share of SUV everywhere sincethe 1980s, are quite illustrative in that respect. Another example is increasing worldwide competitionin the airline industry, resulting in increasing travel and emissions.

    1

  • patents filed in the first period and 150,000 in the second. A firm’s exposure is defined

    as a weighted average of country-level measures of the corresponding variable, where

    the weights proxy for the importance of the various countries to the firm. For compe-

    tition, we also construct a firm-level, Lerner-type index. We find a significant positive

    effect of pro-environment attitudes on the probability for a firm to patent relatively

    more in the clean direction, and this effect is stronger, the higher competition is. In

    particular, our empirical analysis suggests that the combination of realistic increases

    in prosocial attitudes and in product market competition can have the same effect on

    green innovation as a 34% increase in fuel prices worldwide.

    Our research contributes to several literatures. The first one is that on competition

    and innovation (Aghion et al., 1997, 2001, 2005; Vives, 2008). The second is that on

    growth and the environment pioneered by Nordhaus (1994), 3particularly the work on

    endogenous directed technical change analyzing how R&D is shaped by public policies

    such as carbon taxes and/or subsidies to green innovation (Newell et al., 1999; Popp,

    2002; Acemoglu et al., 2012; Aghion et al., 2016). We connect these two literatures

    and bring in individuals’ willingness to “do their part” through their own consumption

    choices, which becomes essential when policy-making is deficient. Third is the liter-

    ature on individual and corporate social responsibility (CSR), both reflecting a mix

    of intrinsic and reputational motivations (Bénabou and Tirole, 2010, 2011; Hart and

    Zingales, 2017); we introduce here product competition as a channel through which

    consumers’ social preferences influence firms’ investment decisions. This also relates

    the paper to experiments such as Falk and Szech (2013) and especially Bartling et al.

    (2015), where lab subjects compete in the roles of both consumers and producers.

    On the empirical side, a number of papers have examined how competition affects CSR

    performance, finding mixed results.4 We depart from this literature in several ways.

    First, we focus on the environmental dimension rather than overall CSR, on the au-

    tomobile industry, and on firms’ innovation decisions rather than their production or

    emissions (which, the model shows, need not go in the same direction). Most impor-

    tantly, we emphasize the interaction, in each firm’s set of markets, between competition

    and consumers’ environmental concerns. Differences in national preferences and firms’

    differential exposures to them not only have a significant effect per se, but turn out

    to be what makes competition actually matter for whether R&D is clean or dirty.

    3See also Nordhaus (2002), Stern and Stern (2007) and Weitzman (2007, 2009)4See Fisman et al. (2006) Fernández-Kranz and Santaló (2010), Flammer (2015), Hawn and Kang

    (2013), and Duanmu et al. (2018).

    2

  • 2 Model

    Time is discrete, with individuals and firms living for one period. At the beginning

    of each period t, firms choose R&D investments, aiming to maximize expected profits.

    Once innovations have realized, firms produce with their respective technologies, com-

    peting for consumers. Revenues are paid out as wages to production and R&D workers,

    and net profits are redistributed to consumers, who are also firms’ shareholders.

    2.1 Preferences

    There is a continuum of differentiated goods, j ∈ [0, 1]. Within and/or across thesesectors, firms potentially differ both by the price they charge and the environmental

    (un)friendliness of the goods they produce. The production or consumption of one unit

    of good with environmental quality q generates x = 1/q units of polluting emissions.

    The representative consumer has standard taste-for-variety preferences, but is also

    concerned about his environmental footprint. When buying yj,f units of quality qj,f

    from each firm f in sector j (denote that set as Fj) , he achieves consumption utility

    Ut =

    ∫ 10

    ln ȳjt dj, (1)

    where

    ȳj =

    ∫f∈Fj

    yj,f (qj,f )δ df (2)

    is his emissions-impact-discounted consumption of variety j. The disutility suffered

    from total emissions will come in subtraction when analyzing welfare, but is taken by

    each individual as given.

    These preferences embody a form of ethical motivation. An individual’s contribution to

    aggregate emissions is negligible, and for instance does not affect the quality of the air

    anyone breathes; nonetheless, he intrinsically dislikes contributing to the externality.

    He feels guilty, or/and socially embarrassed, about the carbon he emits when driving

    or flying, and so would pay a premium for cleaner goods. The parameter δ captures

    the strength of these social-responsibility concerns.

    While sectors are imperfect substitutes, within each one firms’ quality-adjusted offer-

    ings are perfect substitutes. Therefore, all demand for variety j will go to the firm(s) in

    Fj with the highest price/quality ratio, q/p. Furthermore, with logarithmic preferences

    3

  • the same amount will be spent on each variety; we normalize it to 1, choosing current

    expenditure as the numeraire.

    2.2 Technology and market structure

    Labor is the only input, with agents offering an infinitely elastic supply at a wage

    normalized to 1. It takes c units of labor to produce one unit of output (e.g., one car),

    with the firm’s technology determining the associated emissions, 1/q. That technology,

    in turn, reflects the cumulative number kf ∈ N of (green) innovations it made in thepast, or copied from someone who did:

    qf = γkf ,

    where γ > 1 measures the size of a leading-edge environmental innovation. Since

    consumers value a quantity-quality combination (y, q) as yqδ, it effectively takes cγ−δkf

    units of labor for a firm at level kf to produce one unit of quality-adjusted output.

    Suppose that each sector j consists of a duopoly, f = A,B, plus a lagging competitive

    fringe, as follows. First, in each period t both firms have free access to the frontier

    technology achieved in period t − 1. These strong knowledge spillovers simplify theR&D problem, by limiting the investment horizon to a single period.

    Second, a firm’s R&D effort can result in at most one innovation over the current

    frontier: for any z ≤ 1, investing κz2/2 units of labor yields a probability z of inventinga technology that is γ times cleaner, and a probability 1− z of zero progress.

    Together, these assumptions imply that the gap that can open between firms is at most

    one innovation, |kB − kA| ∈ {0, 1}, and it resets to zero at the start of every period.

    A third simplifying assumption is that, at the innovation stage (where kA = kB), only

    one (either) of the two firms has an opportunity to invest in R&D. The other lacks,

    in the current period, a suitable idea or managerial capacity, effectively making its κ

    prohibitively large.

    There can thus, at any point in time, be only be two kinds of sectors: leveled, where

    the duopolists’ qualities are “neck-and-neck”, and unleveled, where a leader is one step

    ahead of its follower. At the start of each period t, which corresponds to the investment

    phase, all sectors are neck and neck, while during the subsequent production phase of

    that period, a fraction z are unleveled, corresponding to the R&D intensity chosen by

    investing firms.

    4

  • In each sector, there is also a competitive fringe of potential entrants. These firms will

    neither produce nor do research in equilibrium but act as a threat, disciplining the

    duopolists. We thus assume that, at the start of each period t, the fringe can costlessly

    imitate the previous-best technology, meaning one that embodies only the k′ = k − 1previous innovations, where k = kA = kB is the level from which the duopolists start,

    and may further innovate.

    2.3 Competition and profits

    Recall that consumers spend the same amount on each variety, and firms in each sector

    compete for that fixed revenue, normalized to 1. Consider first an unleveled sector,

    where an innovation just occurred. The leader has a quality advantage of γδ over

    the follower –its cars pollute γ times less– so it can engage in limit pricing, charging

    pM = γδc and capturing all demand. The output and operating profits of such a de

    facto monopolist are

    yM =1

    pM=

    1

    γδc, πM = 1−

    1

    γδ. (3)

    Consider now a leveled sector, where no innovation recently occurred. If the two firms

    engage in unfettered competition the equilibrium price falls to c, resulting in zero

    profits. Conversely, if they collude perfectly to maximize joint profits, they set p = pM

    like the leader in an unleveled sector, and reap πM/2 each. Indeed, cγδ is the price that

    just keeps out the competitive fringe, which produces goods γ times more polluting

    than those of the duopolists.

    Following Aghion et al. (2005), we span the range between these two extremes by rep-

    resenting (inverse) market competition as the extent to which neck-and-neck firms are

    able to collude at the production-and-sales stage. Thus, we assume that the normalized

    profit for each firm is:

    πD(∆) ≡ (1−∆) πM ,

    where ∆ ∈ [1/2, 1] parametrizes the degree of competition.5 The corresponding priceand sectoral output are given by equating total profits to total sales minus costs:

    5We assume that collusion occurs only at the (ex-post) stage of production and pricing, and notat the ex-ante stage of R&D, which for instance could be harder to monitor.

    5

  • p(∆) =c

    1− 2 (1−∆) πM=

    c

    1− 2 (1−∆) (1− γ−δ)∈ [c, pM ], (4)

    y(∆) =1

    p(∆)=

    1

    c

    [1− 2 (1−∆) (1− γ−δ)

    ]∈[yM ,

    1

    c

    ]. (5)

    For given technologies, competition has the standard effect of forcing down the equi-

    librium price, which increases consumer demand and production. More units produced

    and sold, in turn, result in more emissions –the mass-consumption effect. The other

    consequence of competition is to affect incentives to innovate, which we examine next.

    2.4 Escaping competition through clean innovation

    Recall that each sector starts the current period with both firms neck and neck, then

    one of the two (at random) is endowed with an opportunity for engaging in R&D. If it

    invests z ≤ 1, it succeeds in developing a cleaner technology with probability z, reapingπM ; with probability 1 − z it fails and must collude with its equally able competitor,reaping only πD. A potential innovator thus solves

    maxz∈[0,1]

    {zπM + (1− z) πD(∆)− κz2/2

    },

    resulting in z = min {(πM − πD(∆))/κ, 1}. We restrict attention to parameter valuessuch that

    κ > πM = 1−1

    γδ≡ κ1, (6)

    meaning that innovations are not too easy in terms of their importance or cost. The

    optimal R&D intensity is then always interior,

    z(∆) =∆πMκ

    =∆

    κ

    (1− 1

    γδ

    ). (7)

    Averaging across sectors j ∈ [0, 1], the rate of R&D is also the proportion of themwhere innovation will occur, so the aggregate flow of clean innovations per period is

    simply I ≡ z(∆).

    Proposition 1. Both market competition and consumers’ social-responsibility concerns

    raise investment in, and the total flow of, clean innovations. Moreover, these two forces

    act as complements:∂I

    ∂∆> 0,

    ∂I

    ∂δ> 0,

    ∂2I

    ∂∆∂δ> 0. (8)

    6

  • In a more general model with clean and dirty innovations (e.g., SUV’s), greater com-

    petition would generally enhance both types, but the proportion of clean ones would

    still rise with prosocial values and their interaction with market competition.

    2.5 Pollution and Welfare

    At the production stage of each period, there is a fraction z of sectors in which one

    firm has become cleaner than the other by a factor γ, and a fraction 1 − z where theinnovation effort has failed, so that both still use period t − 1’s frontier technology.Total emissions (normalized by total expenditure) thus equal:

    X = [1− z(∆)] y(∆) + z(∆)yM/γ. (9)

    This is a concave quadratic polynomial in ∆, reflecting two opposing effects. On the

    one hand, by increasing output y(∆) in neck-and-neck sectors, competition directly

    increases pollution. On the other hand, the fear of lower profits causes firms to seek a

    quality advantage through R&D; as a result, a greater fraction z(∆) of sectors develop

    clean technologies, which tends to reduce emissions.

    Proposition 2. Define κ2 ≡ 1 − γ−δ (1 + 1/γ) /2 > κ1 and let κ > κ1, so that theoptimal z(∆) is always interior. As competition ∆ ∈ [1/2, 1] increases:(a) for κ < κ2 − κ1/2, aggregate pollution X(∆) decreases monotonically;(b) for κ > κ2 + κ1/2, X(∆) increases monotonically;

    (c) for κ ∈ (κ2−κ1/2, κ2 +κ1/2), X(∆) is hump-shaped; moreover, it is minimized at∆ = 1 (versus ∆ = 1/2) if and only if κ < κ2;

    (d) for all κ in [κ1, κ2], X(∆) is minimized at ∆ = 1.

    This proposition and the next are illustrated in Figure 1, and proved in the Appendix.

    Proposition 3. Aggregate pollution X(∆) decreases with consumer’s social-responsibility

    concern δ. For all κ > κ1 (more generally, as long as R&D effort is interior) it decreases

    more, the stronger is market competition: ∂2X/∂∆∂δ < 0.

    Let us now evaluate net social welfare. Its first component is utility from consuming

    the z “greener” and the 1− z “dirtier” varieties,

    U = (1− z(∆)) ln y(∆) + z(∆) ln[γδyM ]. (10)

    7

  • Competition raises U through both a quantity effect (higher y(∆)) and a quality effect

    (higher z(∆), reallocating consumption toward cleaner varieties). The second compo-

    nent of wellbeing is environmental quality. Assuming a linear disutility from aggregate

    pollution, welfare equals6

    W = U − ψX, ψ > 0. (11)

    Proposition 2 showed that, when innovation costs κ are relatively high, or competition

    ∆ relatively weak, ∂X/∂∆ > 0. Whether greater competition improves or damages

    social welfare then hinges on how large ψ is. When κ is low and ∆ sufficiently high,

    conversely, ∂X/∂∆ < 0, so ∂W/∂∆ > 0.

    The impact of prosocial concerns similarly depends on how costly R&D is, and on

    the competitive pressure on firms to bear those costs. For fixed z, a higher δ means

    that consumers experience more “guilt” from each unit of pollution embodied in their

    consumption, lowering U. A more environmentally responsible population, however,

    pushes firms to produce cleaner goods: z increases, raising U and lowering X. We show

    in the Appendix:

    Proposition 4. (a) For κ ∈ [κ1, κ2− κ1/2], social welfare W increases monotonicallywith competition; more generally, there is κ̂ > κ2 such that, for all κ ∈ [κ1, κ̂] , W ismaximized at ∆ = 1; (b) W increases with consumers’ environmental concerns δ if and

    only if competition is strong enough. (c) For κ ≥ 2κ1, preferences and competition arecomplements, ∂2W/∂∆∂δ > 0.

    3 Empirical Strategy

    We now test the model’s key predictions for innovation, stated in Proposition 1. Specifi-

    cally, we relate the extent to which a firm innovates in the clean direction to its exposure

    to environmental values and competition, by running regressions of the form:

    Innovationj,t = αV aluesj,t + βCompetitionj,t + γV aluesj,t × Competitionj,t+ δXj,t + Jj + Tt + εj,t (12)

    6These are the only two terms, since: (i) the disutility of labor employed in production and researchis exactly compensated by wage payments; (ii) wages plus operating profits are entirely consumed byindividuals, so that total income equals total spending.

    8

  • In our preferred specification, Innovationj,t is the number of clean patents that firm j

    filed in period t, relative to dirty ones, measured as log(1+number of clean patents) −log(1+number of dirty patents). We also examine clean and dirty patents separately.

    The Jj are firm fixed effects, and the Tt period fixed effects for t = 1998-2002 or 2008-

    2012. We restrict the analysis to these two periods because of data constraints (see

    below).

    V aluesj,t is a firm-specific measure of exposure to pro-environmental values, defined

    as a weighted average of country-level measures, valuesc,t:

    V aluesj,t =41∑c=1

    ωj,c × valuesc,t,

    where ωj,c measures the importance of country c for firm j. In theory one would use

    firms’ sales or profits, but such data is not available. Instead, we compute ωj,c using the

    share of patents filed in country c by firm j between 1950 and 1995, based on the idea

    that protecting intellectual property is more worthwhile where one expects its market

    to be larger. Aghion et al. (2016) show that these weights are very correlated with sales

    for the firms for which country-level sales data is available. We restrict attention to the

    41 countries for which we have data on both environmental values and competition.

    Our main competition measure for firm j in period t is similarly defined as a weighted

    average of country-level indicators. For a subset of firms, we also use a firm-level

    measure that can be interpreted as a Lerner index (see Appendix C). Finally, the Xj,t

    are controls, including GDP per capita, population, (tax-inclusive) oil prices and, in

    some specifications, environmental policies. They are again defined for each firm as a

    weighted average of country-level variables, with weights computed as above.

    4 Data

    4.1 Innovation

    Our innovation measures come from patents in the car industry. Compared to R&D

    investment, patents are available at a more disaggregated level and can thus be classi-

    fied as clean or dirty. Moreover, the auto sector is an innovation-intensive one where

    patents are perceived as an effective means of protection against imitation, something

    not true in all sectors (Cohen et al., 2000). Any given innovation is typically patented

    9

  • in multiple countries, but the European Patent Office’s PATSTAT database allows us

    to track all individual patents belonging to the same patent family. A patent family

    identifies an inventive step that is subsequently patented several times with different

    patent offices. We use this to count families rather than patents, and refer to a family

    as an innovation.

    To classify innovations, we use the International Patent Classification system (IPC)

    and the Y02 classification introduced by the European Patent Office in 2002 to rate

    the climate impact of innovations (both pre- and post-2002). Clean innovations are

    those involving non-fossil-fuel-based propulsion, such as electric or hydrogen cars and

    affiliated technologies (e.g. batteries), while dirty ones are those related to the internal-

    combustion engine (ICE). We label as “grey” technologies those that improve the

    efficiency of the ICE, and as “other” all car-related innovations that do not fit into

    these three categories (see Table C.1).

    Figure 2 shows the worldwide evolution of car-related innovations since the 1960s. The

    annual number has grown from around 3,000 in the 1960s to over 40,000 in 2010.

    Until 2000, this growth was mostly driven by patents in the “other” category, but

    since then clean patents also grew very rapidly. Our sample consists of all firms in the

    industry that patented at least once during either 1998-2002 or 2008-2012.7 This yields

    8,562 firms, out of which 2,130 patented in both periods. In 1998-2002, conditional on

    patenting, the average number of innovations per firm is 2.3 clean ones and 6.1 dirty

    ones; in 2008-2012, these figures are respectively 6 and 3.7. On average, firm-level

    growth rates between the two periods are 34% for clean patents 4% and for dirty ones.

    4.2 Environmental values

    The data on attitudes comes from the International Social Survey Program (ISSP)

    and the World Value Survey (WVS). Several questions could capture the values we

    are interested in, but they are often asked only in a limited set of countries during

    a single survey wave. Only one question is common to both surveys, allowing us to

    cover many countries for two time periods. In the ISSP, it is: How willing would you

    be to pay much higher taxes in order to protect the environment?; and in the WVS,

    Can you tell me whether you strongly agree, agree, disagree or strongly disagree with

    the following statement: ‘I would agree to an increase in taxes if the extra money were

    7Our environmental willingness-to-pay measures are available only during these two periods. Wethus take five-year windows centered on 2000 and 2010, and sum a firm’s annual patents over each.

    10

  • used to prevent environmental pollution’. In both cases, answers are given on a 5-point

    scale.

    Because taxes pertain to public policy more directly than to consumer decisions, we

    also use one additional variable from each survey to create a synthetic index. For

    ISSP, the question is: How willing would you be to pay much higher prices in order to

    protect the environment? For the WVS, it is about (dis)agreement with the statement:

    I would give part of my income if I were certain that the money would be used to

    prevent environmental pollution. We code all answers so that higher values mean more

    pro-environmental attitudes (see Appendix C). We then average all variables at the

    country-period level, transform them into z-scores, and average across all variables

    available for the country-period observation. We thus have data on environmental

    willingness-to-pay for 41 countries for 2 periods, namely 2000 and 2010.

    In most countries, pro-environmental values decreased over this period. This is not a

    specificity of the datasets we are using, nor of the exact point in time when we measure

    attitudes. Appendix B Figure A1 provides a time-series plot of answers to a similar

    question, asked by the Gallup survey to US respondents. The prevailing trend from the

    early 1990s to the beginning of the 2010 decade was a sharp reduction in environmental

    concerns. The reasons for this are unclear, and there is even little awareness of this fact

    in the literature. Figure A1 also shows a sharp reversal after our period of analysis.

    Although we do not have such recent data for other countries, we hypothesize that this

    might be a more general trend. Therefore, in the last section, we will forecast what

    our estimates would imply for green innovation if the decrease in environmental values

    during the first decade of the 2000s was totally erased by their more recent upturn.

    4.3 Competition

    Our main country-level indicator is the World Bank’s openness measure, defined as

    (Imports + Exports)/GDP. We also use, for robustness checks, the Product Market

    Regulation (PMR) indicator from the OECD (Koske et al., 2015), which is available

    only for a subset of countries and years. Both are again translated into z-scores.

    To compute a more direct firm-level measure, we rely on a Lerner-Index-style approach,

    derived from a structural production-function regression. Compared to a standard

    Lerner Index, it allows for non-constant returns to scale and quasi-fixed production

    factors (see Appendix C). This approach requires using balance-sheet data from OR-

    11

  • BIS, and the merge with our patent data is only possible for a subset of firms. This

    firm-level measure for the automobile sector displays much less heterogeneity in trends

    than country-level indicators: most automobile firms experienced a reduction in market

    power during that time period (see Appendix, Figure A2).

    4.4 Country-level controls

    We control for end-user, tax-inclusive automotive fuel prices from the International

    Energy Agency (IEA), real GDP per capita from the World Bank, and population

    from the IMF’s World Economic Outlook. In some robustness checks we also include

    the Environmental Policy Stringency (EPS) Index from the OECD, which provides

    a comprehensive measure of environment-related regulations, taxes, tariffs and R&D

    subsidies. All country-level indicators are transformed into firm-level variables through

    the same weighting approach as for the main regressors.

    4.5 Patent portfolio weights

    Our benchmark definition of country-firm weights ωc,t is the share of a firm’s patents

    filed in each country between 1950 and 1990. The denominator includes not only clean

    or dirty patents, nor automobile-related ones, but include all patents of the firm in the

    relevant countries. Our other weights definitions yield similar results. Whatever the

    definition, the US has the largest weight, on average between 7 and 16%, followed by

    Germany, Japan, the UK and France.

    5 Empirical results

    Table 1 reports our benchmark results, with all magnitudes expressed as z-scores. Panel

    A displays the main effects of environmental values and competition on the direction

    of innovation. Panel B adds an interaction between values and competition. Column 1

    shows the main outcome of interest, namely the growth rate of clean innovation relative

    to dirty ones; Columns 2 and 3 report the effects on both types separately. Finally, in

    Columns 4 and 5, the outcomes are respectively grey innovation and all the “other”

    car-related innovations not classified as either clean, dirty or grey.

    We see that greener consumer values push innovation in the clean direction, primarily

    by reducing the growth rate of dirty patents. Competition has a strong significant

    12

  • positive effect on both types of innovation; although it is stronger for clean than dirty,

    the difference is not significant. Panel B shows that the interaction between values and

    competition has a significant positive effect on the growth rate of clean innovations,

    both in absolute terms (Column 2) and relative to dirty innovations (Column 1).

    Thus, a one-standard-deviation increase in exposure to pro-environmental values is

    associated with a growth rate of clean patents 14% higher than that of dirty patents,

    at the mean level of competition. This effect increases to 17% for levels of competition

    one standard deviation higher than the mean. Predictably, an increase in fuel prices is

    also associated with a higher growth rate of clean patents relative to dirty ones.

    Panel C reproduces Panel B, but controlling for environmental policies. The EPS index

    is only available for 25 countries, so we recompute the weights with this smaller set. As

    was already the case in Aghion et al. (2016), environmental policies do not appear to

    be a significant determinant of innovations: the coefficient on EPS is never significant.

    Including it makes competition’s effect on all types of innovations besides grey ones

    become insignificant, however.

    Table 2 examines the results’ robustness. Panel A explores various weights definitions.

    In Column 1, we incorporate pre-period GDP’s to the weights, based on the idea that

    large countries matter more. Following Dechezleprêtre et al. (2019), we use (GDP ).35:

    if larger markets attract more firms, each firms’ share will decline with country size.8

    The weights become:

    ωj,c =ωj,c ×GDP .35c,pre−period∑41c=1 ωj,c ×GDP .35c,pre−period

    . (13)

    More than half of firms in our sample did not patent in the relevant set of countries

    during the pre-period. In our baseline specification, we assign them uniform weights,

    by adding 1 to the number of patents of a firm in each country. This ensures a smooth

    transition between firms with and without pre-sample patents. In Column 2, we do

    not do this transformation. In Column 3, we drop firms that did not patent in the pre-

    period. In Column 4, we assign them, for each country, the average weight among firms

    that did patent in the pre-period. Results are very consistent across specifications, as

    well as when restricting attention to only car-related patents, or to those with at least

    one citation (available upon request).

    Panel B shows robustness to alternative measures of competition or values. Column

    8Eaton et al. (2011) estimate an elasticity of firms’ average exports to GDP of destination countryof 0.35

    13

  • 1 is the benchmark, identical to Column 1 of Table 1, Panel C. In Column 2 we

    use the “higher tax” question only, instead of our index, to proxy for environmental

    willingness-to-pay. In Column 3 we use the firm-level Lerner measure of competition,

    and in Column 4 the OECD Product Market Regulation measure. Results are robust,

    except in the last specification where the interaction term loses significance.

    Panel C shows robustness to alternative treatments of the “grey” and “other” patents,

    which were dropped in our baseline. “Grey” refers to innovations that make the ICE

    cleaner, hence are neither perfectly clean nor totally dirty; “other” are innovations that

    arguably could be classified as “dirty”. In Column 1, “grey” is included in the “dirty”

    category, in Column 3 with the “clean” one instead. In Column 2, dirty consists of

    dirty, grey and other patents while in Column 4 it consists of dirty and other. These

    changes do not affect results much, except when “grey” is classified as “clean”.

    To summarize, in line with the model’s predictions: pro-environmental values push

    innovation in the clean direction, all the more so when competition is more vigorous.

    Competition per se fosters both types of innovation, with a small but insignificant

    advantage towards cleaner ones.

    6 Accounting and counterfactual exercises

    To examine how economically relevant the effects estimated are, we use our fitted model

    (Table 1, Panel B) to conduct both retrospective and prospective simulations.

    Table 3 (Panel A, Column 1) shows that, between 1998-2002 and 2008-2012, the share

    of clean innovations increased by 23.4 percentage points, while that of dirty innovations

    decreased by 20. How can this be reconciled with the previously mentioned fact that

    citizens in our sample countries generally became less concerned with environmental

    priorities between 2000 and 2010?

    The answer is twofold. First, over that period there was a quintupling of tax-inclusive

    fuel prices, which naturally induces substitution towards cleaner vehicles. More inter-

    estingly, environmental attitudes evolved very differently across countries. If the only

    change had been a uniform decline (the observed mean), the clean share would have

    fallen by 1.0 percentage point, the grey one by 1.5, and the dirty one would have risen

    by 2.4. Because of correlation between firms’s changes in exposure ∆Vj and their level

    of patenting activity (see Appendix B for details), the impact of the properly weighted

    average of ∆Vj’s is somewhat different, but still adverse: Column 2 shows that, evalu-

    14

  • ated at the (patent-weighted) average level of competition C̄, it equals −2.2, 1.2 and 0.8points respectively. Column 3 reveals, however, that values decreased less, or increased

    more, in countries with relatively stronger competition, and especially so for the firms

    that account for the most patenting: this interaction effect dominates the previous

    one, with contributions of 2.9, 1.9, and −4.9 points respectively. The actual impact ofchanges in values was thus positive for the “greenness” of R&D (e.g., 2.9 − 2.2 = 0.7instead of −1.0 for the clean share), in spite of their average decline.

    Similar but smaller compositional effects are present for changes in competition, ∆Cj.

    On average, and evaluated at the average level of environmental values V̄ , they account

    for a rise of 3.7 points for clean patents and a decline of 5.2 for dirty ones (Column

    3); their correlation with environmental concerns augments these numbers by 0.4 and

    −0.3 respectively (Column 5). When all linear and nonlinear effects of preferences andcompetition are included (Column 7), the changes add up to +4.4 for clean and −5.5for dirty. Column 8, finally, incorporates variations in oil prices; the grand total (+29

    for clean and −26.7 for dirty) exceed the observed changes, meaning that other factors(e.g., the Great Recession) must have dampened the shift towards clean.

    In Panel B we turn to a prospective scenario, asking what would happen if –starting

    from the 2008-2012 values– there was an increase in both competition and prosocial

    attitudes. To simulate realistic magnitudes, we use the average absolute changes seen

    between Period 1 and 2. For prosocial values there was a decrease of 0.74 standard

    deviations, and we now simulate a uniform increase of the same size; for competition

    there was an increase of 0.91 standard deviations, and we consider a same-sized uniform

    increase.

    We find that the envisioned increase in prosocial attitudes would raise the share of

    clean innovations by 2.8− 1.0 = 1.8 points, while that in competition would raise it by1.2 + 0.9 = 2.1. Their combined effect is a 4.3 point increase, which is equivalent to

    that of a 34% world-wide rise in fuel prices. Given the often dramatic public reactions

    to even moderate attempts to increase fuel prices (e.g. the French “Gilet Jaunes”),

    this suggests that grassroots and public campaigns to promote citizens’ environmental

    responsibility could be a viable alternative policy option, especially when combined

    with more competitive markets.

    15

  • 7 Conclusion

    Are citizens’ often-stated desires to adopt more environmentally responsible behaviors

    just “cheap talk”, or powerful motivations that end up having a major influence on

    what new products will be developed and sold? And what is the role of market com-

    petition in the process? To answer these questions, we proposed a simple model and

    brought together data on firm-level automotive-sector patents, national environmental

    attitudes, and competition intensity. We found support for the predictions that pro-

    environment attitudes and its interaction with competition both have a significantly

    positive effect on the probability for a firm to aim at cleaner patents. Our results

    are robust to various indicators for environmental values, policies, and product market

    competition.

    More generally, the results provide support for models in which intrinsically or reputa-

    tionally motivated individuals incur costs to act in a “socially responsible” manner in

    spite of having a negligible impact on the aggregate outcome, such as pollution. When

    further leveraged by strong competition between firms, moreover, such prosocial moti-

    vations can actually “move markets”, even at the upstream stage of product research

    and development.

    16

  • References

    D. Acemoglu, P. Aghion, L. Bursztyn, and D. Hemous. The Environment and Directed

    Technical Change. American Economic Review, 102(1):131–166, Feb. 2012.

    P. Aghion, C. Harris, and J. Vickers. Competition and growth with step-by-step inno-

    vation: An example. European Economic Review, 41(3):771 – 782, 1997. Paper and

    Proceedings of the Eleventh Annual Congress of the European Economic Associa-

    tion.

    P. Aghion, C. Harris, P. Howitt, and J. Vickers. Competition, Imitation and Growth

    with Step-by-Step Innovation. The Review of Economic Studies, 68(3):467–492, 2001.

    Publisher: [Oxford University Press, Review of Economic Studies, Ltd.].

    P. Aghion, N. Bloom, R. Blundell, R. Griffith, and P. Howitt. Competition and Inno-

    vation: an Inverted-U Relationship. The Quarterly Journal of Economics, 120(2):

    701–728, May 2005. Publisher: Oxford Academic.

    P. Aghion, A. Dechezlepretre, D. Hemous, R. Martin, and J. Van Reenen. Carbon

    taxes, path dependency, and directed technical change: Evidence from the auto

    industry. Journal of Political Economy, 124(1):1–51, 2016.

    B. Bartling, R. A. Weber, and L. Yao. Do Markets Erode Social Responsibility?

    The Quarterly Journal of Economics, 130(1):219–266, Feb. 2015. Publisher: Oxford

    Academic.

    R. Bénabou and J. Tirole. Individual and Corporate Social Responsibility. Economica,

    77(305):1–19, 2010.

    R. Bénabou and J. Tirole. Laws and Norms. Working Paper 17579, National Bureau

    of Economic Research, Nov. 2011. Series: Working Paper Series.

    W. M. Cohen, R. R. Nelson, and J. P. Walsh. Protecting Their Intellectual Assets:

    Appropriability Conditions and Why U.S. Manufacturing Firms Patent (or Not).

    Working Paper 7552, National Bureau of Economic Research, Feb. 2000. Series:

    Working Paper Series.

    A. Dechezleprêtre, D. Hémous, M. Olsen, and C. Zanella. Automating Labor: Evidence

    From Firm-Level Patent Data. SSRN Scholarly Paper ID 3508783, Social Science

    17

  • Research Network, Rochester, NY, Dec. 2019.

    J.-L. Duanmu, M. Bu, and R. Pittman. Does market competition dampen environ-

    mental performance? Evidence from China. Strategic Management Journal, 39(11):

    3006–3030, 2018.

    J. Eaton, S. Kortum, and F. Kramarz. An Anatomy of International Trade: Evidence

    From French Firms. Econometrica, 79(5):1453–1498, 2011.

    A. Falk and N. Szech. Morals and Markets. Science, 340(6133):707–711, May 2013.

    Publisher: American Association for the Advancement of Science Section: Research

    Article.

    D. Fernández-Kranz and J. Santaló. When Necessity Becomes a Virtue: The Effect

    of Product Market Competition on Corporate Social Responsibility. Journal of

    Economics & Management Strategy, 19(2):453–487, 2010.

    R. Fisman, G. Heal, and V. Nair. A model of corporate philanthropy. Columbia

    University and University of Pennsylvania, 2006.

    C. Flammer. Does product market competition foster corporate social responsibility?

    Evidence from trade liberalization. Strategic Management Journal, 36(10):1469–

    1485, 2015.

    O. Hart and L. Zingales. Companies Should Maximize Shareholder Welfare Not Market

    Value. Journal of Law, Finance, and Accounting, 2(2):247–274, 2017.

    O. Hawn and H. G. Kang. The market for corporate social responsibility (CSR): How

    industry structure determines CSR. Available at SSRN 2380641, 2013.

    I. Koske, I. Wanner, R. Bitetti, and O. Barbiero. The 2013 update of the OECD’s

    database on product market regulation: Policy insights for OECD and non-OECD

    countries. Mar. 2015. Publisher: OECD.

    R. G. Newell, A. B. Jaffe, and R. N. Stavins. The Induced Innovation Hypothesis and

    Energy-Saving Technological Change. The Quarterly Journal of Economics, 114(3):

    941–975, Aug. 1999. Publisher: Oxford Academic.

    W. D. Nordhaus. Managing the global commons: the economics of climate change,

    18

  • volume 31. MIT press Cambridge, MA, 1994.

    W. D. Nordhaus. Modeling induced innovation in climate-change policy. Technological

    change and the environment, 9:259–290, 2002.

    D. Popp. Induced Innovation and Energy Prices. American Economic Review, 92(1):

    160–180, Mar. 2002.

    N. Stern and N. H. Stern. The economics of climate change: the Stern review. cam-

    bridge University press, 2007.

    X. Vives. Innovation and Competitive Pressure*. The Journal of Industrial Economics,

    56(3):419–469, 2008.

    M. L. Weitzman. A Review of the Stern Review on the Economics of Climate Change.

    Journal of Economic Literature, 45(3):703–724, Sept. 2007.

    M. L. Weitzman. On Modeling and Interpreting the Economics of Catastrophic Climate

    Change. The Review of Economics and Statistics, 91(1):1–19, Jan. 2009. Publisher:

    MIT Press.

    19

  • Table 1: The effects of Values and Competition on the direction of innovation

    (1) (2) (3) (4) (5)VARIABLES log (1+ #clean) log (1+ #clean) log (1+ #dirty) log (1+ #grey) log (1+ #other)

    - log (1+ #dirty)

    Panel A: Values and Competition main effectsValues 0.107*** 0.00473 -0.103*** -0.0191 -0.136***

    (0.0211) (0.0191) (0.0179) (0.0157) (0.0239)Competition 0.269 0.514*** 0.246* 0.381*** 0.555***

    (0.166) (0.144) (0.128) (0.108) (0.162)Log fuel price 0.965*** 0.784*** -0.181 -0.0386 0.603***

    (0.156) (0.138) (0.127) (0.114) (0.161)

    Observations 17,124 17,124 17,124 17,124 17,124R-squared 0.122 0.179 0.026 0.052 0.050Number of firms 8,562 8,562 8,562 8,562 8,562

    Panel B: Adding interaction term between Values and CompetitionValues 0.141*** 0.0350 -0.106*** -0.0276 -0.0859***

    (0.0270) (0.0230) (0.0225) (0.0200) (0.0289)Competition 0.167 0.422*** 0.255** 0.406*** 0.403**

    (0.165) (0.140) (0.126) (0.107) (0.161)Values×Comp 0.0296** 0.0268** -0.00278 -0.00750 0.0441***

    (0.0136) (0.0116) (0.0110) (0.00994) (0.0139)Log fuel price 0.596*** 0.450*** -0.146 0.0549 0.0527

    (0.171) (0.149) (0.154) (0.140) (0.215)

    Observations 17,124 17,124 17,124 17,124 17,124R-squared 0.123 0.180 0.026 0.052 0.053Number of firms 8,562 8,562 8,562 8,562 8,562

    Panel C: Robustness to controlling for environmental policiesValues 0.109*** 0.00174 -0.107*** -0.0474*** -0.112***

    (0.0242) (0.0197) (0.0213) (0.0181) (0.0294)Competition -0.0123 0.265 0.277 0.435*** 0.378

    (0.215) (0.174) (0.201) (0.161) (0.264)ValuesXComp 0.0224** 0.0231** 0.000670 -0.00718 0.0376***

    (0.0106) (0.00924) (0.00879) (0.00793) (0.0116)Log fuel price 0.559*** 0.245* -0.314** -0.0648 -0.234

    (0.167) (0.144) (0.148) (0.130) (0.210)EPS 0.235 0.161 -0.0743 0.0615 -0.237

    (0.146) (0.120) (0.138) (0.111) (0.198)

    Observations 17,124 17,124 17,124 17,124 17,124R-squared 0.121 0.180 0.025 0.052 0.050Number of firms 8,562 8,562 8,562 8,562 8,562

    Note: All variables are normalized to z-scores. In panel A and B we use 41 countries to computethe values and competition exposure measures. In panel C, we use only 25 countries, the ones forwhich the Environmental Policy Stringency (EPS) index is available. Besides the coefficients shown,all specifications control for log of population and log of GDP and include firm fixed effects and aperiod fixed effect. Standard errors are clustered at the firm level.

    20

  • Table 2: The effects of Values and Competition on the direction of innovation

    (1) (2) (3) (4)VARIABLES log (1+ #clean) - log (1+ #dirty)

    Panel A: Robustness to different weightsValues 0.117*** 0.0418** 0.168*** 0.266***

    (0.0247) (0.0177) (0.0314) (0.0663)Competition 0.181 0.0747 0.173 0.194

    (0.126) (0.0703) (0.181) (0.428)Values×Competition 0.0391*** 0.0525*** 0.0430** 0.0563***

    (0.0120) (0.0130) (0.0191) (0.0190)Log fuel price 0.427*** 0.0998 0.643*** 0.836**

    ( (0.128) (0.0647) (0.186) (0.369)

    Observations 17,124 17,124 6,704 17,124R-squared 0.120 0.102 0.176 0.122Number of firms 8,562 8,562 3,352 8,562

    Panel B: Robustness to different values and competition measuresValues 0.109*** 0.153*** 0.177*** 0.102***

    (0.0242) (0.0500) (0.0421) (0.0250)Competition -0.0123 0.283 0.00387 0.00949

    (0.215) (0.223) (0.0319) (0.124)Values×Competition 0.0224** 0.0546** 0.0658* 0.0162

    (0.0106) (0.0217) (0.0352) (0.0161)Log fuel price 0.559*** 0.402 1.416* 0.722***

    (0.167) (0.269) (0.731) (0.258)EPS 0.235 0.398*** 0.124 0.328**

    (0.146) (0.150) (0.256) (0.148)Competition measure World Bank World Bank Lerner OECDValues measure Index Higher Tax Index IndexObservations 17,124 17,124 2,706 17,124R-squared 0.121 0.120 0.199 0.120Number of firms 8,562 8,562 1,854 8,562

    Panel C: Robustness to different treatments of ”other” and ”grey”Values 0.175*** 0.188*** 0.0590** 0.176***

    (0.0283) (0.0205) (0.0245) (0.0198)Competition -0.210 -0.210 0.498*** -0.111

    (0.182) (0.140) (0.170) (0.132)ValuesXCompetition 0.0407*** 0.0177** 0.00756 0.0125

    (0.0142) (0.00787) (0.0118) (0.00765)Log fuel price 0.413** -0.0456 0.645*** 0.0291

    (0.196) (0.164) (0.160) (0.160)

    Clean clean clean clean + grey cleanDirty dirty + grey dirty+ grey + other dirty dirty + other

    Observations 17,124 49,482 17,124 49,482R-squared 0.070 0.051 0.149 0.050Number of firms 8,562 24,741 8,562 24,741

    Note: All variables are normalized to z-scores. Besides the coefficients shown, all specifications controlfor log of population and log of GDP and include firm fixed effects and a period fixed effect.

    21

  • Table 3: Historical and Prospective Counterfactuals

    Share (1) (2) (3) (4) (5) (6) (7) (8) (9)change Actual C————————————— Due To —————————————–B

    5(α+ γC̄

    )×∆V

    γ(C − C̄

    )×∆V

    (β + γV̄

    )×∆C

    γ(V − V̄

    )×∆C

    γ∆V×∆C ∆V,∆C

    ∆V,∆C,∆P

    Other

    Panel A: HistoricalClean: 23.8 -2.2 3.0 2.9 1.2 -0.5 4.4 32.5 -8.7Grey: -3.2 1.2 2.0 1.9 -0.8 0.2 1.1 -5.9 2.7Dirty: -20.7 0.8 -1.2 -4.9 -0.5 0.1 -5.5 -26.7 6.0

    Panel B: ProspectiveClean: 2.8 -1.0 1.2 0.9 0.7 4.3 - -Equiv. ∆P/P (%) 21 0 9 6 5 34 - -

    Note: Share changes are in percentage points. Column 1 reports historical evolutions; Columns 2and 3, those due solely to changes in (firms’ market exposures to) environmental values ∆Vj and theirinteractions with ( exposures) to competition levels Cj , the average of which is C̄; Columns 4 and 5 dothe same for changes ∆Cj in competition and their interactions with value levels Vj . Column 6 givesthe “second order” effects from interactions between the ∆Vj and ∆Cj . Column 8 computes the totalchanges attributable to variations in values, competition, and oil prices. See Appendix B for details.

    22

  • Figure 1: Effect of competition and social values on pollution

    Figure 2: Evolution over time of clean, dirty, grey and other car related innovations

    (a) Absolute number of innovations

    050

    0010

    000

    1500

    020

    000

    2500

    0

    1960 1980 2000 2020Year

    Grey DirtyClean Other

    by categoryNumber of car related patents

    (b) Relative share

    0.2

    .4.6

    .8

    1960 1980 2000 2020Year

    Share of clean Share of dirtyShare of grey Share of other

    Among all car related patentsShare of clean, dirty, grey and other patents

    Source: PATSTAT. Patents classified as clean, dirty, grey or other based on the IPC and Y02classification systems. See main text for more details.

    23

  • ONLINE APPENDIX

    Appendix A: Proofs

    Proof or Proposition 2. For all κ > κ1, so that z is interior, we can write total

    emissions (9) as:

    X(∆) =

    (1− ∆πM

    κ

    )[1− 2(1−∆)πM ] +

    κγπM(1− πM). (A.1)

    Focusing first on the extremes of full competition and full collusion to get the main

    intuitions, the former is less polluting than the latter if X(1) < X(1/2), or

    1− πMκ

    +πM(1− πM)

    κγ<

    (1− πM

    )(1− πM) +

    πM(1− πM)2κγ

    ⇐⇒

    πM(1− πM)2κγ

    <(

    1− πM2κ

    )(1− πM)−

    (1− πM

    κ

    )=πM(1 + πM)

    2κ− πM ,

    which simplifies to

    κ < 1− γ−δ

    2

    (1 +

    1

    γ

    )= κ2, (A.2)

    where κ2 > 1 − γ−δ = πM = κ1 was first defined in Proposition 2. Quite intuitively,for any given κ, (A.2) holds when γ or/and δ is large enough. Let us next determine

    where X achieves its maximum on [1/2, 1] :

    κ∂X

    ∂∆= −4π2M∆ + (2κ− 1 + 2πM) πM +

    1

    γπM(1− πM), (A.3)

    so ∂X/∂∆ > 0 if and only if

    ∆ <1

    4πM

    (2κ− 1 + 2πM +

    1− πMγ

    )=

    1

    2+

    1

    4πM

    (2κ− 1 + γ

    −δ

    γ

    )≡ ∆̂X(κ, γ, δ).

    (A.4)

    Naturally, ∆̂X is increasing in κ and decreasing in both γ and δ. Moreover,

    ∆̂X(γ, δ) <1

    2⇐⇒ κ < 1

    2

    (1− γ

    −δ

    γ

    )= κ2 −

    πM2≡ κ3, (A.5)

    ∆̂X(γ, δ) > 1 ⇐⇒ κ > κ3 + πM = κ2 +πM2≡ κ4 (A.6)

    where κ2 > κ1 = πM was first defined in Proposition, by equation (A.2).

    24

  • It then follows that (maintaining κ > κ1, thus ensuring an interior optimum for z) :

    (i) If κ < κ2 − κ1/2, Z is decreasing in ∆, and thus minimized at ∆ = 1.(ii) If κ > κ2 + κ1/2, then Z is increasing in ∆, and thus minimized at ∆ = 1/2.

    (iii) If κ ∈ (κ2 − κ1/2, κ2 + κ1/2) then X is hump-shaped in ∆, with a maximum at∆̂Z(γ, δ) ∈ (1/2, 1) and a minimum either at 1/2 or at 1, depending on κ ≷ κ2 (recallthat this is what defines κ2).

    Note, finally, that conditions κ > πM and κ < κ2 − πM/2 define a nonempty intervalwhen 3πM < 2κ2, that is, γ

    −δ (2− 1/γ) > 1, or

    δ < ln (2− 1/γ) / ln γ. � (A.7)

    Proof or Proposition 3. From (A.1), when κ > κ1, we have

    κ∂X

    ∂πM=

    κ

    [−1 + 2(1−∆)πM +

    1− πMγ

    ]− 2(1−∆)

    (1− ∆πM

    κ

    )− ∆πM

    κγ. (A.8)

    The last two terms are clearly negative, and so is the first, since (1− πM) /γ < 1−πM ≤1−2(1−∆)πM for all ∆ ≥ 1/2. Recalling that πM = 1−γ−δ, it follows that ∂X/∂δ < 0.When κ ≤ κ1, R&D effort may be (depending on ∆) at a corner, z = 1, in which caseX = yM/γ = 1/cγ

    −δ−1, which decreases in δ. Finally, differentiating (A.3) in πM ,

    κ∂2X

    ∂∆∂πM= −4(1− 2πM)∆ +

    1

    γ(1− 2πM)− 2κ+ 2(1− 2πM)− 1

    = (1− 2πM)[

    1

    γ+ 2− 4∆

    ]− 1− 2κ.

    If 1 − 2πM ≥ 0, the right-hand side is bounded above by (1 − 2πM)/γ − 1 − 2κ <1/γ− 1− 2κ < 0. If 1− 2πM < 0, it is bounded above by (2πM − 1) (2− 1/γ)− 1− 2κ,since ∆ ≤ 1; but πM ≤ 1, so this expression is at most 1 − 1/γ − 2κ < 0, sinceκ > κ1 = πM = 1 − 1/γδ > 1 − 1/γ. Therefore, ∂2X/∂∆∂δ < 0 for all ∆, as long asκ > κ1.

    Proof or Proposition 4. Part (a). This follows from the conjunction of ∂X/∂∆ < 0

    for κ ≺ κ2 − κ1/2 , by Proposition 2, and

    ∂U

    ∂∆=πMκ

    ln

    (1

    1− 2(1−∆)πM

    )+

    (1− ∆πM

    κ

    )2πM

    1− 2(1−∆)πM> 0. (A.9)

    25

  • Part (b). Recalling (3), (5) and (10), we can rewrite

    U =

    (1− ∆πM

    κ

    )ln (1− 2(1−∆)πM) + ln

    (1

    c

    ), (A.10)

    ∂U

    ∂πM=

    κln

    (1

    1− 2(1−∆)πM

    )− 2(1−∆)πM

    1− 2(1−∆)πM

    (1− ∆πM

    κ

    ), (A.11)

    Thus, ∂U/∂πM > 0 if and only if

    κ < ∆

    [πM + f

    (2(1−∆)πM

    1− 2(1−∆)πM

    )], (A.12)

    where f(t) ≡ ln(1 + t)/t for all t > 0 and f(0) ≡ limt→0 f(t) = 1. Note that f is adecreasing function, since f ′(t) has the sign of g(t) ≡ t− (1 + t) ln(1 + t), where clearlyg′(t) < 0 = g(0) for all t > 0. The right-hand side of (A.12) is thus increasing in ∆, so

    the inequality holds if and only if ∆ > ∆(πM , κ), with

    ∆(πM , κ) < 1 ⇐⇒ κ < 1 + πM , (A.13)

    ∆(πM , κ) < 1/2 ⇐⇒ κ <1

    2

    [πM + f

    (πM

    1− πM

    )]≡ κ̄(πM), (A.14)

    Condition (A.13) is always compatible with κ > πM and κ < κ2 − πM/2. Condition(A.14), which ensures that ∂U/∂πM > 0 for all values of ∆ ∈ [1/2, 1], is more demand-ing since κ̄(πM) < (1 + πM) /2 and compatible with κ > πM , only if

    πM < f

    (πM

    1− πM

    )=

    ln [1/(1− πM)]πM/(1− πM)

    ⇐⇒ π2M < (1− πM) ln(

    1

    1− πM

    ), (A.15)

    which holds for instance when πM is small enough, meaning that δ ln γ is small enough.

    This finishes to establish (b).

    Part (c). In (A.9), the first term is increasing in πM , and while the second not always

    is, a sufficient condition is that (∆πM/κ) (1−∆πM/κ) be increasing, which occurs for∆πM/κ < 1/2; conversely, πM/κ < 1/2 is necessary the second term for that same

    term to be increasing in ∆ up to ∆ = 1. Thus, when κ > 2πM = 2κ1, we have

    ∂2U/∂∆∂δ > 0.

    We check, finally, that this new lower bound on κ is compatible with key upper bounds

    previously defined, meaning that they jointly define a nonempty set of values for

    (κ, γ, δ). We have:

    26

  • 2πM < κ2 − πM/2 ⇐⇒ 5(1− γ−δ) < 2κ2 = 2− γ−δ(1 + 1/γ) ⇐⇒

    δ <ln(4/3− 1/3γ).

    ln γ. (A.16)

    2πM < κ̄(πM) ⇐⇒ 3πM < f(

    πM1− πM

    )=

    ln [1/(1− πM)]πM/(1− πM)

    ⇐⇒ 3π2M < (1− πM) ln(

    1

    1− πM

    ). (A.17)

    The first condition is naturally tighter than (A.7), so when it holds we have ∂2U/∂∆∂δ >

    0 for all ∆ and ∂U/∂δ > 0 for ∆ in some nonempty interval (∆, 1]. If the second

    condition also holds (which is ensured by some additional upper bound on δ), then

    ∂2U/∂∆∂δ > 0 > ∂U/∂δ > 0 for all ∆ ∈ [1/2, 1]. This, together with the fact that,from Proposition 2, ∂2X/∂∆∂δ ≺ 0 for all κ � κ1, establishes Part (c). �

    Appendix B: Counterfactual Methodology

    We can write our regression model (12 )as

    Zjt ≡ ln (PATjt + 1) = αV jt + βCjt + γVjt × Cjt + ϕFjt + εjt, (B.1)

    where, for each firm j and time, PATjt is the number of patents (families) of a given

    type (clean, dirty, etc.), Vjt and Cjt are its (average) degrees of exposure to prosocial

    values and competition respectively, and Fjt collects all other explanatory variables,

    such as oil prices, firm and period fixed effects, etc. There is one such estimation

    for each patent type, but for simplicity we abstain here from indexing the regression

    coefficients by “clean,” “dirty”, etc.

    Denoting ∆Xjt = Xjt − Xjτ any historical or counterfactual change between dates τand t, and given estimated coefficients (α̂, β̂, γ̂, ϕ̂), the implied patenting level at t is

    P̂AT jt = (PATjτ + 1)× exp(∆̂Zj)− 1, (B.2)

    where (omitting time subscripts to lighten the notation):

    ∆̂Zj ≡ α̂∆Vj + β̂∆Cj + γ̂(∆Vj ×Cj) + γ̂(Vj ×∆Cj) + γ̂(∆Vj ×∆Cj) + ϕ̂∆Fj. (B.3)

    For small changes, ∆̂PAT j is proportional to ∆̂Zj, and can thus be decomposed into

    27

  • the constituents of (B.3). Alternatively, one can use the fitted nonlinear model for

    counterfactual analysis, asking: “How much would total of patents of each type have

    increased or decreased between τ and t, if the only changing factor had been the vari-

    ations in environmental values observed in the different countries, and thus firms’

    exposures V jt? Or, replacing historical accounting by prospective simulations: “How

    much should we expect those patent numbers to increase between τ and (some future)

    t, if the only changing factor will be some assumed set of ∆V ’s (or/and ∆C’s)?

    The answer is obtained by setting, for each j, all terms in (B.3) to zero except for

    α̂∆V j + γ̂(∆Vj × Cj), then summing across firms the resulting ∆̂PAT j’s computedfrom (B.2). This total change can itself be attributed to the combination of a direct,

    “average” effect of the ∆Vj’s (weighted by initial patenting activity), and one that

    reflects their interaction, and therefore their correlation pattern, with initial levels of

    competition, Cj. This is again clearest when understood as a first-order approximation,

    ∆̂PAT ≡∑j

    ∆̂PAT j ≈∑j

    (PATj + 1) ∆̂Zj

    = α̂∑j

    (PATj + 1) ∆Vj + γ̂∑j

    (PATj + 1)Cj ×∆Vj

    =(α̂ + γ̂C̄

    )∑j

    (PATj + 1) ∆Vj + γ̂∑j

    (PATj + 1)(Cj − C̄

    )∆Vj,(B.4)

    where C̄ ≡ (1/N)∑

    j (PATj + 1)Cj is the average level of (firm exposure to) compe-

    tition, with each firm weighted by its initial patenting activity.9 Alternatively, to get

    exact numbers we can simulate the nonlinear model, by:

    (a) Setting, for all j, all changes in (B.3) except ∆Vj to zero, and equating all Cj’s

    to C̄; the results for clean, grey and dirty patents are given in Column 2 of Table 3.

    They correspond to what would have happened if every firm had faced the (patent-

    weighted) average attitudinal change, and the (patent-weighted) average level of market

    competition.

    (b) Setting all terms but the γ̂(∆VjCj)’s to zero, and subtracting γ̂C̄∑

    j (PATj + 1)×∆Vj. This yields the results in Column 4, reflecting the (patent-weighted) extent to

    which firms that saw larger ∆Vj’s in their markets were exposed there to higher or

    lower levels of competition. As seen in Tables B.1 and B.2, between τ = 2002 and

    9We use the patent levels corresponding to the “clean” category. Using those for “dirty” or “grey”instead, or using each one for the corresponding version of (B.2)-(B.4), leads to broadly similar results.

    28

  • t = 2012, firms’ ∆Vj’s were overall sightly positively correlated with their Cj − C̄’s,but strongly so where the ∆Vj’s were most important (top percentile), and especially

    among firms most active in patenting activity. The weighted covariance of the two

    variables is thus 1.63 (versus an unweighted one of 0.002), almost twice as large as the

    weighted-mean effect of −0.91. That is why, in Panel A of Table 3, Column 3 showspositive contributions of changing values to clean and negative to dirty, which swamp

    the adverse mean effects from Column 2.

    Similarly, Columns 4 and 5 in Panel A compute the counterfactual changes in each

    number of patents (relative to total) corresponding to historical changes in competition

    only, doing so separately for the effect of the (patent-weighted) average change, evalu-

    ated at the mean level of environmental values,(β̂ + γ̂V̄

    )∑j (PATj + 1)×∆Cj, and

    that reflecting the correlation pattern with initial attitudes, γ̂∑

    j (PATj + 1)(Vj − V̄

    ∆Cj,, where V̄ ≡ (1/N)∑

    j (PATj + 1)Vj.

    Column 7 incorporates all the above effects, plus those of the interaction in changes,

    γ̂(∆Vjt×∆Cjt). Column 8, finally, simulates the full fitted model (B.1)-(B.2), in whichthe ϕ̂∆Fjt’s include in particular includes variations in oil prices.

    The prospective exercise reported in Panel B of Table 3 is identical, except that the

    initial date is τ = 2012 and the counterfactual ∆Vjt’s and ∆Cjt’s are taken to be

    uniform across firms, equal respectively to 0.74 and 0.86 standard deviations. As

    explained in Section 6 (see also Table B.1), these magnitudes are the historical ones

    observed in our sample, but with a sign reversal for the former –in line with the

    fact that, since 2012 (when our patent dataset ends), the previous general decline

    environmental values seems to have given way to an upswing.

    29

  • Table B.1: Descriptive Statistics for Counterfactual Calculations

    Unweighted Patent-WeightedMean Std.Dev. P1 P50 P99 Mean Std.Dev.

    ∆V alues -0.742 0.766 -2.357 -0.911 2.227 -0.909 1.250∆Comp 0.861 0.160 0.347 0.915 1.145 0.760 0.223

    (Comp− Comp)×∆V alues 0.002 1.407 -2.125 -0.379 6.747 1.631 3.117(V alues− V alues

    )×∆Comp -0.062 0.722 -2.635 -0.124 1.908 0.395 1.229

    ∆V alues×∆Comp -0.609 0.896 -1.38 -0.834 2.389 -0.542 1.393∆log(FuelPrice) 1.698 0.202 1.601 1.601 2.381 1.963 0.343

    Note: Values and Competition are measured as in Table 1, Panel B. Patent weighting is defined inequation B.4, using firms’ clean patent levels in 2002.

    Table B.2: Correlations between key variables (2008-2012)

    Clean Grey Dirty Values CompetitionClean 1Grey 0.869 1Dirty 0.443 0.659 1Values 0.106 0.084 0.017 1Competition -0.154 -0.154 -0.108 -0.604 1∆Values -.028 0.015 0.103 -0.489 0.003 1∆Competition -0.81 -0.066 -0.013 -0.403 0.680 0.246 1

    Note: Clean, Grey and Dirty correspond here to (one plus) each firms’ number of patents in eachcategory, in 2002. The measures of Values and Competition used are the same as in Table 1, Panel B.

    30

  • Appendix C: Details on variable definition

    C1. Classifying patents as clean, dirty or grey

    Table C.1 reports the Cooperative Patent Classification (CPC) classification used to

    determine the different flavours of innovation.10

    Table C.1: Patent CPC classification codes used

    CleanY02T10/60 Other road transportation technologies with climate change mitigation effect

    Y02T10/70 Energy storage for electromobility

    Y02T90/10 Technologies related to electric vehicle charging

    Y02T90/34 Fuel cell powered electric vehicles

    Y02T90/42 Hydrogen as fuel for road transportation

    GreyY02T10/10 Climate change mitigation technologies related to fuel injection

    Y02T10/20 Climate change mitigation technologies related to exhaust after treatment

    Y02T10/40 Climate change mitigation technologies related to engine Management Systems

    Y02T10/50 Climate change mitigation technologies related to Intelligent Control Systems

    DirtyF02 Combustion Engines

    Other AutomotiveB60 Vehicles in General

    C2. Values

    Answers to the ISSP question “How willing would you be to pay much higher taxes in

    order to protect the environment?” vary from 1 (‘very willing’) to 5 (‘very unwilling’)

    and we reverse-code them, so that a higher value means a more pro-environmental

    attitude. In the WVS, answers to the corresponding question are 1 (‘strongly agree’),

    2 (‘agree’), 4 (‘disagree’) and 5 (‘strongly disagree’). We code as 3 the ‘don’t know’

    answers and reverse-code the others, as for the ISSP. Answers to the two additional

    questions we use from WVS and ISSP are again reverse coded to ensure consistency.

    Our data cover most major economies, and in particular most countries in which firms

    innovating in the automotive sector reside, with a few notable exceptions such as Italy

    and Spain.

    10See https://www.cooperativepatentclassification.org/index, as well as also https://www.wipo.int/classifications/ipc/en/ and https://www.epo.org/news-issues/issues/classification/classification.html.

    31

    https://www.cooperativepatentclassification.org/indexhttps://www.wipo.int/classifications/ipc/en/https://www.wipo.int/classifications/ipc/en/https://www.epo.org/news-issues/issues/classification/classification.htmlhttps://www.epo.org/news-issues/issues/classification/classification.html

  • C3. OECD PMR indicator

    The OECD Product Market Regulation (PMR) indicator is a comprehensive variable

    that aggregates responses from a questionnaire of over 700 questions, falling into three

    main areas: state control, barriers to entrepreneurship, and barriers to trade and in-

    vestment. We use it as robustness and not as our main measure, because it does not

    cover as many countries and years as the World Bank index. The two measures have a

    correlation of 0.3. Indeed, some countries rank very differently along them, like the US

    which is among the least open according to the World Bank, but the most competitive

    besides Great Britain according to the OECD.

    C3. Computation of firm-level Lerner Index

    We estimate firm-level measures of competition using a (revenue) production function

    framework. We assume a homothetic translog production function with materials Mit

    and labor Lit as flexible factors, and capital Kit a quasi-fixed production factor. A

    firm’s (log) revenue (Rit) growth can then be written as

    ∆rit ≈λ

    µ̄it+ s̄Mit (∆mit −∆kit) + s̄Lit (∆lit −∆kit) +

    1

    µ̄it∆ωit, (C.1)

    where ∆rit = ln(Rit/ lnRit−1) (and equivalently for production factors), λ is a scale

    parameter, s̄Mit = (sMit + sMit−1) /2 the average share of materials expenditure in

    revenue between period t and t − 1 (and equivalently for labor inputs), and ωit aHicks- neutral shifter of TFP or/and demand. µ̄it is the average markup of prices over

    marginal cost between period t and t−1, making µ̄it −1 a Lerner index specific to firmi at time t. Short run profit maximization implies

    sMit =αMitµit

    , (C.2)

    where αMit is the elasticity of output with respect to changes in production factor M

    (and analogously for labor). Note that in the translog case,

    αMit = αM + αKMkit + αLM lit + αMMmit. (C.3)

    This specification is consistent with a wide variety of market structures. For further

    discussion see Martin (2012) and Forlani (2016). We can rewrite (C.1) as

    32

  • ΞitᾱMitλ−∆kit =

    1

    λ∆ωit, (C.4)

    where

    Ξit ≡∆rit − λµ̄it + s̄Mit (∆mit −∆kit) + s̄Lit (∆lit −∆kit)

    s̄Mit.

    Given assumptions on the evolution of the ∆ωit shock, we can fit this to firm-level data

    using a GMM approach. Thus, if ∆ωit follows an AR(1) process, ωit = ρωit−1 + ηit

    where ηit is iid, we can write

    η̂it = ΞitᾱMitλ−∆kit −

    ρ

    λ

    [Ξit−1

    ᾱMit−1λ

    −∆kit−1],

    and estimate the parameters δ = [ρ/λ, αM/λ, αKM/λ, αLM/λ, αMM/λ] using the mo-

    ment conditions:

    E

    [η̂it ×

    {Ξit−1,

    1

    ∆kit,k̄it

    ∆kit,l̄it

    ∆kit,m̄it∆kit

    }]= 0.

    After identifying δ, we can compute α̂Mit/λ using (C.3). Then, from (C.2) we can

    compute

    λ̂

    µit= sMit

    (α̂Mitλ

    )−1, (C.5)

    which is an inverse Lerner Index, scaled by the returns to scale parameter λ; i.e. it tells

    us the excess of markups over returns to scale. While this is different from the markup

    over marginal costs, it is more relevant in terms of measuring market power, as revealed

    by excess earnings over what would be reasonable to compensate for increasing returns.

    We also implement a simpler version, assuming a Cobb Douglas production function,

    so that αMit = αM . Both approaches lead to similar results.

    Note that these firm-level measures, focusing specifically on the automobile sector,

    display much less heterogeneity in trends than the country-level indicators. Panel (a)

    of Figure A2 shows deciles of the distribution of markups over marginal costs – i.e.,

    the inverse of the Lerner Index – across firms. It indicates that markups (and thus

    competition) have been flatlining over time, with the exception of the top decile, where

    we see an upward trend from 2003 onwards. Panel (b) shows changes in market power

    for continuing firms between 2002 and 2012: for the majority of automobile firms, the

    general picture is that of a reduction in market power during that time period.

    33

  • 34

  • Appendix Figures

    Figure A1: Long run decline and recent reversal in pro-environmental concerns

    Source: “Preference for Environment Over Economy Largest Since 2000”, by Lydia Saad forGallup News, April 2019

    Figure A2: Firm-level Markups

    (a) Distribution over time

    .91

    1.1

    1.2

    1.3

    1.4

    1995 2000 2005 2010 2015year

    (b) Change between the 2 periods

    0.2

    .4.6

    .8D

    ensi

    ty

    -2 -1 0 1 2Change in Markups

    Notes: Panel (a) shows centiles (10th to 90th percentile) of firm-level markups (inverseof the Lerner index) over time. Panel (b) shows the distribution of changes in markupsbetween 2002 and 2012. These markups are computed using ORBIS data.

    35

    1 Introduction2 Model2.1 Preferences2.2 Technology and market structure2.3 Competition and profits2.4 Escaping competition through clean innovation2.5 Pollution and Welfare

    3 Empirical Strategy4 Data4.1 Innovation 4.2 Environmental values4.3 Competition4.4 Country-level controls4.5 Patent portfolio weights

    5 Empirical results6 Accounting and counterfactual exercises7 Conclusion Appendix A: Proofs Appendix B: Counterfactual Methodology Appendix C: Details on variable definition